MemoryHelm
Award 1 in the Agents for Good track of the AI Agents: Intensive Vibe Coding Capstone Project, a Google Featured Hackathon on Kaggle.
Competition recognition; not an endorsement or clinical validation.

Clear daily guidance for people living with dementia—with caregivers in control.

MemoryHelm helps caregivers turn everyday activities into clear, accessible routines delivered through visual and voice guidance. Every routine requires caregiver review and approval before it becomes active.

Watch the demo Explore the project
Open-source prototype · Synthetic demonstration data · Not clinically validated
TODAY / 09:30A
Good morning, Anna☼ 18°

NEXT ROUTINE

Water the plants

09:30
Safety checkedCaregiver approved
  1. 1

    Take the small watering can.

  2. 2

    Fill it halfway with water.

  3. 3

    Give each plant a little water.

THE GUIDING PRINCIPLE

The caregiver remains at the helm.

AI helps structure, simplify and review each routine. A trusted person decides whether it should become active. Technology provides support, while people retain control.

A DAY WITH MEMORYHELM

See how MemoryHelm could support a caregiver through an everyday routine.

This short illustrative scenario follows a caregiver using structured visual and voice guidance while retaining responsibility for every decision.

Playing this video connects to YouTube.

Watch on YouTube

Illustrative scenario using synthetic content. Not clinical advice or evidence of effectiveness.

Why it matters

A small amount of structure can protect a lot of independence.

MemoryHelm is designed around a human reality: people may need support with daily routines while still wanting to direct their own day.

57m

people living with dementia worldwide

2021 estimate
10m

new cases every year

Global estimate
$1.3t

estimated global cost

2019 estimate
5h

of care and supervision each day

Average informal caregiver
A growing need for thoughtful support78m 2030→ 139m 2050

Projected number of people living with dementia worldwide. These figures describe the context — not an outcome MemoryHelm claims to change.

View source
Data sources: World Health Organization
How it works

From a simple request to a supported moment.

01

Describe

A caregiver shares an everyday activity in their own words.

02

Structure

The request becomes a short, predictable sequence of steps.

03

Safety check

Potential risks and unclear instructions are surfaced for review.

04

Human approval

A trusted caregiver edits, reviews and decides what goes live.

05

Guide

The person sees one calm routine with visual and voice guidance.

06

Respond

Completion, a help request or a call keeps the care loop visible.

The product experience

One view for the caregiver. One calm view for the person being supported.

Working prototype
CAREGIVER VIEW

Review before it reaches Anna.

Draft routineReview needed

Water the plants

“Help Anna water the plants after breakfast. Keep it simple and remind her where the small can is.”

AI
Structured draft ready3 steps · Low cognitive load
Safety classification: low risk Audit trail ready
TODAY VIEW

Simple enough to follow.

Good morning, Anna09:30
☼YOUR NEXT ROUTINE

Water the plants

One step at a time, with a voice when you need it.

1 Take the small watering can.2 Fill it halfway with water.3 Give each plant a little water.
Human-centered by design Help is always visible
Safety by design

Useful AI knows when to slow down.

MemoryHelm can organize support, but it does not quietly take over responsibility. Boundaries are part of the product, not a footnote.

MemoryHelm does

  • Turn caregiver requests into clear, short steps
  • Surface potential risks and ambiguity
  • Require explicit caregiver approval
  • Make completion and help requests visible
  • Keep controlled actions traceable

MemoryHelm does not

  • Diagnose dementia
  • Provide medical advice
  • Manage medication
  • Detect or handle emergencies
  • Replace caregivers or healthcare professionals
Under the hood

A multi-agent system with a clear chain of responsibility.

Different agents handle different kinds of work. Their output stays reviewable, bounded and accountable to a human decision.

01

Root Orchestrator

Coordinates the flow and keeps decisions traceable.

02

Routine Planning Agent

Turns everyday language into structured, practical steps.

03

Safety Policy Agent

Checks for risk, ambiguity and missing human review.

04

Dementia-Friendly Communication Agent

Simplifies language and lowers cognitive load.

05

Escalation Agent

Makes requests for help visible to the caregiver.

Built with Google ADK, Model Context Protocol, Next.js, FastAPI and PostgreSQL. Controlled actions are recorded in an audit trail.
MEMORYHELM / ORIGINSupport is a bridge,
not a replacement.
The story

Built from care, shaped by responsibility.

MemoryHelm began from Ventsislav Asparuhov’s personal experience supporting his late grandmother through dementia.

The original prototype, created under the name MemoryBridge, received Award 1 in the “Agents for Good” category of the Google & Kaggle AI Agents Capstone.

The name changed. The promise stayed: help people preserve moments of independence without asking caregivers to give up visibility or control.

Recognition

Recognized by Google & Kaggle in the Agents for Good category

Originally submitted under the name MemoryBridge, the project was selected as the Award 1 winner in the ‘Agents for Good’ category of the Google & Kaggle AI Agents: Intensive Vibe Coding Capstone. The project is now continuing its development under the name MemoryHelm.

The recognition reflected not only the project’s purpose, but also its safety-first architecture and commitment to keeping people in control.

AWARD 1 / AGENTS FOR GOOD
1

MemoryBridge
now MemoryHelm

353,000+Registered course participants
392,000+Active participants on Kaggle Discord
12,000+Active capstone participants
6,000+Submitted capstone projects

MemoryBridge was selected as the Award 1 winner in its category from this global field of projects.

Program statistics reported by Google
Official Winners

View the official Google & Kaggle competition results.

View the winners Project Write-up

Read the original MemoryBridge competition submission and technical overview.

Read the write-up Video Demonstration

See how the caregiver workflow, safety review and assisted-user experience work together.

Watch the video
What comes next

From an awarded prototype to responsible real-world validation.

The next step is not immediate deployment. We are looking to learn from caregivers, dementia professionals and care organizations that can review the workflow and help shape its continued development.

Discuss the project
COLLABORATION WORKSPACE
PRIVATE PILOT · INVITATION-BASED

A private workspace for structured review and responsible progress.

Invited research, healthcare, caregiving, and project collaborators can review MemoryHelm, complete guided evaluations, and keep recommendations documented in one clear workflow.

Open the workspace
How collaboration worksPRIVATE PILOT
Receive an invitation

Access is provided directly to approved project collaborators.

01
Review the application

Explore the current MemoryHelm workflow in a focused environment.

02
Submit structured feedback

Document observations, recommendations, and priorities through guided evaluation forms.

03
Follow responses and progress

Keep feedback, project responses, and subsequent improvements visible in one written process.

04
Research notes

From historical leads to testable questions.

MemoryHelm explores responsible research directions in cognitive health and dementia care. These evidence-informed papers identify questions worth investigating without presenting unproven ideas as established treatments.

First page of Ancient Medical Traditions and Modern Dementia Care
Evidence-informed narrative review

Ancient Medical Traditions and Modern Dementia Care

A critical review of historical leads, current treatments, and responsible research priorities – separating potentially testable ideas from unsupported treatment claims.

By Ventsislav Asparuhov · MemoryHelm Research

Read paper Download PDF
First page of Wuqinxi and Mild Cognitive Impairment
Research concept paper

Wuqinxi and Mild Cognitive Impairment

An evidence-focused hypothesis paper examining whether a structured motor-cognitive practice merits careful pilot research in people with mild cognitive impairment.

By Ventsislav Asparuhov · MemoryHelm Research

Read paper Download PDF

MemoryHelm research papers are educational and exploratory. They have not been peer reviewed and are not clinical guidelines, medical advice, or substitutes for professional diagnosis or care.

The people behind it

A thoughtful project needs thoughtful collaboration.

VA

Ventsislav Asparuhov

Creator, Project Lead & Lead Developer

Leads MemoryHelm’s product vision, safety-first architecture, AI-agent design, and technical development.

LinkedInmail@memoryhelm.com
KP

Kevin Pinard

Project Collaborator

Supports MemoryHelm through outreach, partnership development, and connections with research institutions, healthcare organizations, and caregiving communities.

PortfolioACT Eustis